RESEARCH Open Access

Continuous sweep versus discrete stepprotocols for studying effects of wearablerobot assistance magnitudePhilippe Malcolm1,2,3* , Denise Martineli Rossi1,2,4, Christopher Siviy1,2, Sangjun Lee1,2,Brendan Thomas Quinlivan1,2, Martin Grimmer5 and Conor J. Walsh1,2

Abstract

Background: Different groups developed wearable robots for walking assistance, but there is still a need formethods to quickly tune actuation parameters for each robot and population or sometimes even for individualusers. Protocols where parameters are held constant for multiple minutes have traditionally been used forevaluating responses to parameter changes such as metabolic rate or walking symmetry. However, these discreteprotocols are time-consuming. Recently, protocols have been proposed where a parameter is changed in acontinuous way. The aim of the present study was to compare effects of continuously varying assistancemagnitude with a soft exosuit against discrete step conditions.

Methods: Seven participants walked on a treadmill wearing a soft exosuit that assists plantarflexion and hip flexion.In Continuous-up, peak exosuit ankle moment linearly increased from approximately 0 to 38% of biological momentover 10 min. Continuous-down was the opposite. In Discrete, participants underwent five periods of 5 min withsteady peak moment levels distributed over the same range as Continuous-up and Continuous-down. We calculatedmetabolic rate for the entire Continuous-up and Continuous-down conditions and the last 2 min of each Discreteforce level. We compared kinematics, kinetics and metabolic rate between conditions by curve fitting versus peakmoment.

Results: Reduction in metabolic rate compared to Powered-off was smaller in Continuous-up than in Continuous-down at most peak moment levels, due to physiological dynamics causing metabolic measurements in Continuous-up and Continuous-down to lag behind the values expected during steady-state testing. When evaluating theaverage slope of metabolic reduction over the entire peak moment range there was no significant differencebetween Continuous-down and Discrete. Attempting to correct the lag in metabolics by taking the average ofContinuous-up and Continuous-down removed all significant differences versus Discrete. For kinematic and kineticparameters, there were no differences between all conditions.

Conclusions: The finding that there were no differences in biomechanical parameters between all conditionssuggests that biomechanical parameters can be recorded with the shortest protocol condition (i.e. singleContinuous directions). The shorter time and higher resolution data of continuous sweep protocols hold promise forthe future study of human interaction with wearable robots.

Keywords: Exosuit, Protocol, Parameter sweep, Metabolic, Kinematics

* Correspondence: pmalcolm@unomaha.edu1John A. Paulson School of Engineering and Applied Sciences, HarvardUniversity, Cambridge, MA 02138, USA2Wyss Institute for Biologically Inspired Engineering, Harvard University,Cambridge, MA 02138, USAFull list of author information is available at the end of the article

© The Author(s). 2017 Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, andreproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link tothe Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver(http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.

Malcolm et al. Journal of NeuroEngineering and Rehabilitation (2017) 14:72 DOI 10.1186/s12984-017-0278-2

BackgroundIn recent years, there have been a growing number ofdifferent academic and industry groups developing wear-able robots for walking assistance in able-bodied and clin-ical populations. Typically, devices have been evaluated andoptimized for specific objectives by means of studies wherea device parameter is systematically varied while changes inan objective parameter are measured [1–12]. Differentstudies and devices have targeted different objectives, suchas reducing metabolic rate [1–3, 5, 7, 9, 10, 13, 14], increas-ing walking speed [13], altering muscle activation [6], im-proving perceived comfort [12] and restoring symmetry [9].Given that these studies are examining human-machineinteractions, it is not surprising that results can vary fordifferent wearable robots [4] or populations [13]. Con-sequently, there is a need for methods to evaluate theeffects of wearable robot parameter settings quickly andefficiently [12, 15, 16] for each new each populationand objective and for new types of wearable robotssuch as soft exosuits [2, 3, 7, 11, 17].Initially, most experiments with wearable robots used

discrete step protocols [1–5, 7, 9, 11]. Discrete step proto-cols, also called steady-state mapping [15], involve meas-uring an objective parameter for a certain amount of timeat a number of parameter settings within a range of inter-est. This is followed by a curve-fitting process to identifywhich parameter setting results in the minimum or max-imum objective parameter value. More recently, groupsstarted using continuous sweep protocols [12, 15, 16].Continuous sweep protocols involve continuously evaluat-ing the goal parameter (e.g. metabolic rate, perception, …)while continuously changing the device parameter bysmall increments for every subject step or stride.When focusing specifically on metabolic rate, several

challenges arise in the use of discrete step protocols.Due to noise in indirect calorimetry measurements, thistype of protocol requires averaging a large number ofbreaths [18–20] and sometimes even averaging multipletrials or multiple participants to obtain a reliable esti-mate of the underlying landscape [21, 22]. Due to thedelay between changes in local energetic cost and themetabolic response at the pulmonary level, every step ofa discrete step protocol must run for 2 to 3 min beforesteady-state metabolic rate measurements can start [23].Further, current wearable robots can require severalminutes to adapt to the wearer’s gait after a change incontroller is applied (e.g. [17]). This also increases therequired walking time at each discrete step before thecollection of useful data can start. Consequently, to findthe optimal parameter settings with a sufficiently highconfidence level, participants have to walk for long pe-riods of which the majority is under sub-optimal param-eter settings. Walking for a long time with sub-optimaldevice parameters (e.g. too little or too much assistance)

can be challenging for both the robot (e.g. excessivewear and physical drift of the device on the person [24])and the wearer (e.g. clinical populations may be unableto maintain prolonged effort with suboptimal assistance).Further, high levels of exertion can lead to cardiopulmo-nary drift [21], decreasing the accuracy of conditionscollected late in an experimental session. Thus, requiredtime for data collection, exercise tolerance of the partici-pants and time availability limit the number of parame-ters settings that can be studied with discrete stepprotocols. Limited conditions reduce the resolution bywhich the landscape of metabolic rate versus parameterchanges can be analyzed.In the field of exercise physiology, continuous sweep

protocols have been proposed for identifying cardiopul-monary thresholds and have shown promise as a quickeralternative to discrete step protocols which is especiallyuseful for clinical populations that cannot engage inextended activity [19, 21, 23, 25, 26]. Felt et al. [15]suggested using a continuous sweep protocol foroptimization, and successfully validated this approachagainst a discrete step protocol in an experiment involv-ing subject step frequency optimization. In a continuoussweep protocol, metabolic rate at a certain parametersetting is delayed compared to a discrete step protocol.Felt et al. developed a method that removes this delay bymodeling changes in metabolic rate with a first-order lin-ear dynamic model with a single time constant based onwork by Selinger et al. [22]. Having been successful usedin exercise physiology and studying step frequency, it willbe important to understand if continuous protocols canbe applied to parameter tuning for wearable robots, inparticular given the known complexity of human-exoskeleton interactions [27].The primary focus of the present study was to evaluate

differences in metabolic rate and gait biomechanicsbetween evaluation protocols that use continuous anddiscrete variations of actuation parameters. In a continu-ous sweep protocol, it is known that there is an apparentlag or delay between parameter changes and the corre-sponding changes in metabolic rate. This can be due tothe bulk effect of physiological dynamics and biomech-anical response delay. The apparent delay in metabolicrate is not an absolute delay but is known to follow anexponential evolution over time. Conversely, we definethe time that is required by a trained user to changebiomechanics in response to a parameter change as“biomechanical response delay”.Our specific aims and hypotheses are as follows: (1)

Measure if there is biomechanical response delay incontinuous conditions by comparing biomechanicalmetrics between continuous and discrete conditions.We hypothesize that potential biomechanical responsedelay is small or non-existent when the continuous

Malcolm et al. Journal of NeuroEngineering and Rehabilitation (2017) 14:72 Page 2 of 13

changes in wearable robot parameters are gradual andthus would be easy for participants to adapt to, ex-cept if the parameter change is such that the usercan never adapt to it, such as if the force is too highto allow walking. (2) Evaluate the physiological dy-namics during the continuous conditions by compar-ing reduction in metabolic rate between continuousand discrete conditions. We hypothesize that mea-sured reductions in metabolic rate in continuousconditions will be lagging behind the values thatwould be expected during steady-state testing. In turnwe expect this will result in higher reductions mea-sured in the continuous condition where the peakforce goes from high to low, and lower reductions inthe continuous condition where the peak force goesfrom low to high and the reductions in discreteconditions being halfway between both continuouscondition directions. (3) Evaluate two methods to correctfor physiological dynamics during testing with a continu-ous protocol: taking the average of two sweep directionsof a continuous condition and using the instantaneousmetabolic cost estimation from Selinger et al. [22] and Feltet al. [15]. Building further on our earlier hypothesis thatmetabolic reduction in a discrete sweep is midwaybetween continuous sweep directions we hypothesize thattaking the mean of both directions of a continuous sweepwill closely estimate the results of a discrete sweep. Also,building further on our hypothesis that biomechanical re-sponse delay will be small or non-existent we hypothesizethat the instantaneous cost estimation using an assumedfixed time constant similar to previous literature willclosely estimate metabolic reduction from discrete sweep.To validate the use of continuous and discrete proto-

cols for optimizing wearable robotics we comparedboth protocols in a parameter study with a soft textilebased exoskeleton, called exosuit, developed by ourgroup [2, 3, 7, 11, 17]. In previous studies with otherwearable robots (e.g. [5, 28]), groups have found thathigher assistance magnitude can lead to higher reduc-tions in metabolic rate while other studies in havefound that higher assistance magnitude can also lead toless reduction in metabolic rate (e.g. [5]). Therefore, webelieve it is relevant to do an experiment involvingchanging the assistance magnitude to compare bothparameter study methods for measuring the effect ofpeak assistance force or moment with our soft exosuit.

MethodsParticipantsWe tested seven participants (27 ± 5 yrs.; 68 ± 10 kg;1.7 ± 0.1 m; mean ± SD). The study was approved bythe Harvard Longwood Medical Area InstitutionalReview Board and all experiments were conducted in

accordance with this approved protocol. All participantsprovided written informed consent.

ConditionsParticipants walked on a treadmill at 1.5 m s−1 undertwo conditions while parameters were varied: A Discretestep condition and two Continuous sweep conditions(Continuous-up and Continuous-down).In Discrete, participants underwent a series of 5-min

periods with steady peak forces at zero (Powered-off ),18.7 (Low), 37.5 (Med.), 56.2 (High), and 75.0% bodyweight (Max.). This is similar to how the majority ofstandard parameter studies were conducted in the litera-ture [1–5, 7, 10, 11].In Continuous-up, the desired peak exosuit ankle force

increased from zero to 75% body weight over 10 min.Continuous-down was the opposite. Before each Con-tinuous condition participants walked for 4 min underthe assistance level from the start of the parametersweep and after each Continuous condition they walkedfor another 3 min under the assistance level from theend of the parameter sweep. Assuming a 42-s timeconstant [22] and knowing that most of the change inexoskeleton force happens quickly just after starting theexoskeleton controller, this should allow sufficient timefor metabolic rate to approach the steady state value.

ProtocolBefore the testing session participants did a training sessionthat involved a total of one-hour walking with all the con-ditions that would be repeated during the testing session.During the testing session, participants began with an 8-

min warm-up period during which they briefly experi-enced all force magnitudes before the actual measure-ments began. We randomized the order of the conditionssuch that participants either started by completing all theContinuous conditions or by completing all of the forcelevel of the Discrete condition. Within the Continuousconditions participants alternatively started with eitherContinuous-up or Continuous-down. In Discrete, we alsorandomized the order of the five force levels and groupedthem into two uninterrupted walking blocks, each 15 minin length and each containing a Powered-off trial for rela-tive comparison of metabolic data. Between all testingblocks we gave a 5-min break.

ExosuitThe participants wore a soft exosuit consisting of aspandex base layer, a waist belt, a calf wrap on each leg,and two vertical straps per leg crossing from the back ofthe calf wrap, through the center of the knee joint axis,to the front of the waist belt (Fig. 1). This configurationassists both ankle plantarflexion and hip flexion [11, 17].

Malcolm et al. Journal of NeuroEngineering and Rehabilitation (2017) 14:72 Page 3 of 13

Actuation system and sensorsWe used an offboard actuation system to generate assist-ive forces and Bowden cables to transmit the forces tothe exosuit at the ankle joint. The Bowden cable sheathswere connected to the back of the calf wrap and theinner cables were connected to a metal bracket at theback of the heel of the boot. As the motor retracted thecable, the distance between the two attachment pointswas shortened, generating a tensile force that wasdistributed between the calf wrap and the vertical straps,resulting in a plantarflexion moment around the ankleand flexion moment around the hip [11, 17].We attached one gyroscope (LY3100ALH, STMicroe-

lectronics, Geneva, Switzerland) on each foot to segmentgait cycles and attached one load cell (LTH300, FUTEKAdvanced Sensor Technology, Irvine, CA, USA) on eachcalf wrap to measure the force exerted at the Bowdencable attachment. To measure the assistive force trans-mitted to the hip joint, we attached two load cells(LSB200, FUTEK Advanced Sensor Technology, Irvine,CA, USA) in series with the two vertical straps and thewaist belt.

ControlWe used a triangular motor position profile to deliverassistive moments similar to the biological ankle mo-ment pattern considering both ankle joint kinematicsand human-suit effective stiffness, as described in [11].During each stride, the motor retracted the Bowdencable at a constant rate starting from the heel-strikeuntil the end of the stance phase. Just before the swingphase, the motor released the cable out such that thesystem did not hinder the wearer’s natural motion

during the swing phase. The system performed a force-based position control on a stride-by-stride basis similarto [7], which made decisions on the slope of the triangu-lar trajectory for the next stride by comparing thedesired and the actual peak forces at the end of eachstride. The desired peak force for each stride was varieddepending on the experimental conditions. In theDiscrete conditions, the desired peak force was set as aconstant at each desired force level. On the other hand,in Continuous-up/down conditions, the desired peakforce was set to linearly increase (Continuous-up) ordecrease (Continuous-down), as described in the experi-mental protocol section. In Discrete the controllerrequired about 1 min to reach the desired assistancemagnitude since the controller was programmed togradually adjusted the motor position profile over anumber of steps. In Continues-up and Continuous-downthe controller followed the desired assistance magnitudewith a delay of about 3 s. The controller achieved a meanerror in peak force of 0.027 ± 0.085 N kg−1 and a meanabsolute error in peak force of 0.308 ± 0.049 N kg−1.

MeasurementsWe calculated metabolic rate based on rates of oxygenconsumption and carbon dioxide production [29] mea-sured via indirect calorimetry (K4b2, Cosmed, Rome,Italy). We analyzed metabolic rate of every participant ofContinuous-up and Continuous-down and the last 2 minof each of the force levels of Discrete. We did not usenoise eliminating methods other than averaging theselast 2 min. We calculated the change in metabolic ratebetween each active condition and Powered-off. ForDiscrete we used the Powered-off that we recorded

Load cells

Gyroscopes

Bowden cable(outer sheath)

OffboardActuation system

Bowden cable(inner cable)

Soft exosuit

a b

Fig. 1 Experimental setup. a Participant wearing a soft exosuit, the actuation system (offboard, Bowden cable, sensor locations). Part of this figureis reproduced from [30]. b Frontal view showing the medial and lateral straps running from the front of the waist-belt to the rear ofthe calf-wraps

Malcolm et al. Journal of NeuroEngineering and Rehabilitation (2017) 14:72 Page 4 of 13

within the same 15-min bout of walking. For Continu-ous-up we used the last minute at Powered-off before thestart of Continuous-up and for Continuous-down weused the last minute at Powered-off after the end ofContinuous-down.We measured body segment motions using a motion

capture system (Vicon, Oxford Metrics, Oxford, UK;120 Hz). We collected three-dimensional ground reac-tion forces using an instrumented split-belt treadmill(Bertec, Columbus, OH, USA; 2160 Hz). We filtered allmarker and force signals using a zero-lag fourth orderlow pass Butterworth filter with a 5–9 Hz cut-off fre-quency that we selected using a custom residual analysisalgorithm (MATLAB, MathWorks, Natick, MA, USA).We calculated sagittal plane moment arms of the exosuitat the ankle and hip, joint angles, joint moments, andjoint powers for the right leg using Visual 3D (C-Mo-tion, Rockville, MD, USA). To evaluate biomechanicalresponse delay, we analyzed peak joint angles thatwere previously shown to significantly change with in-creasing peak exosuit ankle moment in [30]. To havea measure of global biomechanical adaptation withsame units as metabolic rate (i.e. W kg−1), we alsocalculated a center-of-mass based metabolic rate esti-mate [4, 5, 31] from the sum of all positive and nega-tive center-of-mass power, minus exosuit power,

multiplied by respective muscle efficiencies from Mar-garia et al. [32].We used an automatic gait event detection algorithm

(Visual3D, C-Motion, Rockville, MD, USA) to determineheel strikes and segment kinematic and kinetic data intogait cycles. From Continuous-up and Continuous-downwe processed the entire 10 min of biomechanical dataand from Discrete we processed the last minute of eachforce level.

Data analysisContinuous-up and Continuous-down each contain a fewhundred data points (metabolic rate for every breath andbiomechanics for every stride) whereas Discrete containsfive data points (averaged data from five force levels). Toallow comparison between both Continuous-up andContinuous-down with Discrete, we had to reorganizethe data prior to running statistics due to this unequalsampling rate.According to the design of Continuous-up and

Continuous-down, stride-by-stride peak exosuit anklemoment varied linearly with time (Fig. 2a, c). Becausemetabolic rate measurements took place breath bybreath, this did not match up temporally with any givenstride’s peak exosuit ankle moment. To accommodatethis, we first calculated a linear fit between peak exosuit

Continuous conditions Discrete conditions

Peak exosuit ankle moment (Nm kg-1)

Peak exosuit ankle moment (Nm kg-1)

Exo

suit

an

kle

mo

men

t (N

m k

g-1)

Pea

k ex

osu

itan

kle

mo

men

t (N

m k

g-1)

Met

abo

lic r

ate

(W

kg

-1)

Met

abo

lic r

ate

(W

kg

-1)

Protocol time (min)

Protocol time (min)

Gait cycle (%)

Protocol time (min)

Protocol time (min)

Gait cycle (%)

a b

c d

e f

g h

Fig. 2 Data organization in Continuous-up, Continuous-down and Discrete based on example data from one participant. a Peak exosuit anklemoment over the gait cycle of Continuous-up and Continuous-down condition. The peak moment values ranged from Powered-off to Maximum(Max.) applied moment (Continuous-up) and from Max. to Powered-off (Continuous-down). b Peak exosuit ankle moment over the gait cycle ofDiscrete. Peak moment values at Powered-off, Low, Med., High, and Max. applied moment at the ankle. c Peak moment values and time forContinuous-up (green line) and Continuous-down (blue line). The thin line shows actual peak moments and the thick line shows curve fit of momentversus time. The order of Continuous-up and Continuous-down was randomized for each participant. d Five minutes at each moment level inDiscrete in a randomized order. The thin line shows actual peak moments and the thick line shows commanded peak moment. e Metabolic rateover time for Continuous-up (blue points) and Continuous-down (green points). Points represent breaths. f Metabolic rate over time of each Discretemoment level. g Metabolic rate fitted at each peak moments level from Powered-off to Max. h Average metabolic rate of the last 2 min at eachmoment level from Powered-off to Max

Malcolm et al. Journal of NeuroEngineering and Rehabilitation (2017) 14:72 Page 5 of 13

ankle moment and experiment time. Evaluating this fitat the time points corresponding to metabolic measure-ments provided a corresponding peak exosuit anklemoment for every metabolic rate measurement inContinuous-up and Continuous-down (Fig. 3).Next, in the Continuous-up and Continuous-down, we fit

breath-by-breath metabolic rate to peak exosuit ankle mo-ment with a second order polynomial (Fig. 2g). In Discrete,we fit the steady-state metabolic rate to peak exosuit anklemoment with a second order polynomial (Fig. 2h). The se-lection of the curve fitting order was based on evaluationof significances of coefficients trying different options fromthird to first order (Additional file 1). We used the samemethods for fitting stride-by-stride biomechanical parame-ters versus peak exosuit ankle moment.

We evaluated two ways of compensating for physio-logical dynamics in the data from Continuous-up andContinuous-down. As a first method, we performed ourcurve fitting on the combined data from Continuous-upand Continuous-down, a method we define as Continu-ous-bidirectional (Fig. 4). As an alternative approach toadjust for physiological dynamics, we also utilized the in-stantaneous cost mapping function outlined by Felt et al.[15] and Selinger et al. [22] (Fig. 5). We applied theirmethods with a fixed assumed time constant of 42 s (thisvalue was chosen based on the average observed timeconstant in [22]) on the section of the data including thelast minute of steady-state data before the continuousparameter change and the 3 min of steady state dataafter the continuous parameter change. The results ofthe instantaneous cost mapping of Continuous-up andContinuous-down were respectively called AdjustedContinuous-up and Adjusted Continuous-down. For allconditions (Discrete, Continuous-up, Continuous-down,Adjusted Continuous-up and Adjusted Continuous-down)we calculated all the curve fits on individual data and thenaveraged to plot the population average. In addition toapplying the instantaneous cost mapping with a fixedassumed time constant of 42 s we also solved for theactual subject time constants that optimally reduced thesum of the squared differences between Continuous-upand Continuous-down, between Continuous-up andDiscrete and between Continuous-down and Discrete andwe reported these average best fitting time constants.

StatisticsFor testing differences between conditions, we first eval-uated the individual curve fits of each condition at thedifferent peak exosuit ankle moment levels of Discrete.We conducted repeated-measures ANOVA to test forsignificant differences between Continuous-up, Continu-ous-down and Discrete for the curve fits evaluated at thedifferent peak moment levels in order to evaluate abso-lute differences between conditions. We also usedrepeated-measures ANOVA to check for differencesbetween conditions for the delta between the curve fitsevaluated at the highest peak moment minus the curvefits evaluated at the lowest peak moment in order toevaluate the effects of different conditions on the relativeparameter changes over the entire peak moment range.Continuous-bidirectional was included in this analysis asa fourth condition. We applied these aforementionedanalyses on metabolic rate, peak joint angles and thecenter-of-mass based metabolic rate estimate.We also separately applied the same analysis to test

for significant differences between Adjusted Continuous-up, Adjusted Continuous-down and Discrete. On mea-sures that showed significant condition effects, we per-formed paired t-tests to compare conditions. We then

Fig. 3 Exosuit ankle moment interpolation. Metabolic rate issampled once per breath. Exosuit ankle peak moment is sampledonce every stride. In order to be able to relate metabolic rate topeak moment we calculated a linear fit between peak exosuit anklemoment and experiment time. This linear fit is then evaluated attime points corresponding to metabolic measurements

Malcolm et al. Journal of NeuroEngineering and Rehabilitation (2017) 14:72 Page 6 of 13

compared Continuous-up versus Continuous-down to as-sess if differences existed that may be due to metabolicor biomechanical response delay. We also comparedContinuous-bidirectional, Continuous-up and Continu-ous-down each versus Discrete to understand which con-dition may be a best match to Discrete. All statisticalanalyses were within-subject so this does not take intoaccount between-subject variability. We used a signifi-cance threshold of α = 0.05.

ResultsExosuit kineticsPeak exosuit ankle moments in Discrete were0.014 ± 0.002, 0.179 ± 0.004, 0.360 ± 0.060,0.544 ± 0.010, 0.699 ± 0.0170 N m kg−1 (mean ± s.e.m.)respectively for the Powered-off, Low, Med., High andMax. Peak exosuit ankle moments during the minutebefore and after Continuous-up were 0.015 ± 0.002and 0.686 ± 0.035 N m kg−1. Peak exosuit ankle

= p < 0.05

Cont.-up vs. Cont.-downCont.-up vs. Discrete

Cont.-upCont.-bidirect.Cont.-downDiscrete

Fig. 4 Change in metabolic rate plotted against exosuit ankle peak moment. Black, green, blue and red line respectively represent populationaverage second order polynomial curve fits for Discrete, Continuous-up, Continuous-down and the average of Continuous-up and Continuous-downcalled Continuous-bidirectional. Shaded areas represent standard error of Discrete, Continuous-up and Continuous-down (for Continuous-up standarderror is only plotted in the positive direction and for Continuous-down it is only plotted in the negative direction). Bi-colored symbols representsignificant differences (P ≤ 0.05) between the curve fits of the different conditions with the colors in the symbol evaluated at peak moment levelsfrom Discrete. Time series plots show population average exosuit ankle moment profiles at the peak moment levels of Discrete

Cont.-up

Adjusted Cont.-upCont.-downDiscreteAdjusted Cont.-down

Fig. 5 Instantaneous metabolic rate estimation. Adjusted Continuous-up and Adjusted Continuous-down based on instantaneous metabolicrate estimation [15, 22] assuming a fixed time constant of 42 s based on [22]. Dashed green and blue line represent population average forinstantaneous metabolic rate estimation respectively for Continuous-up and Continuous-down. Black line represents population average forsecond order polynomial curve fit from Discrete. Bi-colored squares represent significant differences

Malcolm et al. Journal of NeuroEngineering and Rehabilitation (2017) 14:72 Page 7 of 13

moments during the minute before and after Continuous-down were 0.652 ± 0.034 and 0.013 ± 0.002 N m kg−1. Insummary, these metrics show that we were able to sweeppeak exosuit ankle moments over similar ranges in allconditions.

Biomechanical adaptationThere were no significant differences between any of themethods in joint angles and the center-of-mass basedmetabolic rate estimate (all p-values >0.24) (Table 1).However, the center-of-mass based metabolic rate esti-mate showed strong significant correlations with the ac-tual change in metabolic rate (P < 0.001 and R2 = 0.70,Pearson’s correlation test). In summary, these results showthat there were no significant differences in the evaluatedbiomechanical parameters between all conditions.

Metabolic rateMetabolic rate followed a monotonically decreasingtrend versus peak exosuit ankle moment in all theconditions (Fig. 4). Change in metabolic rate in Discretevaried from zero at Powered-off to −1.11 ± 0.14 W kg−1

at Max. Change in metabolic rate in Continuous-upvaried from +0.17 ± 0.12 W kg−1 at the start to−0.70 ± 0.11 W kg−1 at the end. Change in metabolic rate

in Continuous-down ranged from −0.95 ± 0.14 W kg−1 atthe start to +0.15 ± 0.10 W kg−1 at the end. Change inmetabolic rate in Continuous-bidirectional ranged from+0.16 ± 0.11 W kg−1 at Powered-off to −0.83 ± 0.13 W kg−1

at Max.The curve fit of Continuous-up had a significant

positive intercept (P = 0.02). The curve fit of Continu-ous-down also had a large positive offset trending towardsignificance (P = 0.06). Reduction in metabolic rate inContinuous-up was significantly smaller than in Discreteat all peak exosuit moment levels from Low to Max.(P = 0.01, 0.03, 0.007, 0.03). ()The average slope of thereduction in metabolic rate over the tested peak momentrange was significantly smaller in Continuous-up than inDiscrete (P = 0.02). In Continuous-down and Continu-ous-bidirectional the average slopes of the reduction inmetabolic rate over the tested peak moment range werenot different from Discrete (P = 0.98 and 0.33 respect-ively). In summary, the results show that there were largesignificant differences in metabolic rate at the differentpeak moment levels between conditions but the averageslope of the trend was more similar between conditions.After applying the instantaneous metabolic rate

estimation with a fixed time constant of 42 s [22],Adjusted Continuous-up was still higher than Adjusted

Table 1 Biomechanical parameter comparison

Individual curve fits of biomechanical parameters and metabolic rate were evaluated at the peak exosuit moment of Med. for Continuous-up, Discrete, Continuous-bidirectional and Continuous-down Brackets indicate significant differences (P ≤ 0.05). The peak joint angles in the table were selected based on the kinematicalparameters that were shown to significantly increase or decrease with increasing peak exosuit ankle moment in [30]. Center-of-mass based metabolic rateestimate was selected as a global parameter that would represent the entire biomechanical changes and would have a good correlation with metabolic rate ina magnitude variation experiment [5, 28]. For comparison, change in metabolic rate is also reported in the table

Malcolm et al. Journal of NeuroEngineering and Rehabilitation (2017) 14:72 Page 8 of 13

Continuous-down at High (P = 0.02) which indicates thatthe time constant was larger than we expected. AdjustedContinuous-up was higher than Discrete at Powered-offand High (P = 0.04 and 0.02). There was also a trend to-ward a significant difference between the instantaneousmetabolic rate estimate from Adjusted Continuous-upand Discrete at Max. (P = 0.06). This indicates thateven after applying the instantaneous cost estimationContinuous-up and Continuous-down did not closelymatch with Discrete. When we solved for the actualsubject time constants we found that a time constant of106 +/−29 s optimally reduces the sum of the squareddifferences between Continuous-up and Discrete and atime constant of 49 +/− 20 optimally reduces the sumof the squared differences between Continuous-downand Discrete. Additional file 2 shows a simulation ofwhat the difference between conditions would havelooked like if differences would have been exclusivelydue to physiological dynamics following an exponentialchange with a time constant of 42 s [22]. In summary,these results show time constants that are higher thanthe assumed time constant of 42 s and that varydepending on which conditions are tried to match.

DiscussionUnderstanding how to select parameter values for wear-able robots is important for guiding their development.This study examines the differences in biomechanicaland physiological outcomes while varying parametervalues in a continuous (Continuous-up and Continuous-down) and discrete step (Discrete) manner. We variedthe level of exosuit assistance delivered at the ankle jointthrough a multi-articular soft exosuit and comparedbiomechanical and metabolic measurements to thosecollected at steady state conditions at four differentdiscrete assistance levels. In addition, we expected differ-ences between the Continuous-up and Continuous-downconditions due to physiological dynamics and potentialbiomechanical response delay.For the selected kinematic and kinetic parameters that

we evaluated, we found no differences between Continu-ous-up and Discrete and between Continuous-down andDiscrete. Further, we did not find any difference in vari-ability parameters between the Continuous-up andDiscrete and between Continuous-down and Discrete(Additional file 3). This indicates that participants hadno difficulty in adapting to the slow linear increase ordecrease in peak exosuit ankle moment in the Continu-ous-up and Continuous-down conditions. This echoesthe interpretation of a recent treadmill walking study inwhich the authors assumed that participants can per-fectly adapt to treadmill belt speed changes if they areimplemented in a slow linear way similar to Continuous-up and Continuous-down conditions [33]. To check how

important the slow linear increase or decrease in peakexosuit ankle moment in Continuous-up and Continu-ous-down was to avoid biomechanical response delaywe did an additional analysis to check if there was re-sponse delay in biomechanics in the Discrete condi-tions (Additional file 4). Based on average results itappears that there was no biomechanical responsedelay even in the changes between the Discrete forcelevels. It should be noted that in even the Discreteconditions the changes between force levels were slowdue to the limitations of the soft exosuit controller soit is possible that biomechanical response delay wouldhave existed with more abrupt transitions betweenthe Discrete force levels.Since there were no differences between Continuous-

up and Discrete and between Continuous-down andDiscrete, our results indicate that for kinematic or kin-etic measurements Discrete could be replaced by eitheronly Continuous-up or only Continuous-down whichrequired 4 + 10 min for data collection whereas Discreterequired 30 min). It should be kept in mind howeverthat in addition to the actual protocol time participantsalso completed a training day and warm-up that consti-tuted an additional hour and 15 min of walking and thatthe results could also have been different if participantshad different levels of training.Another potential advantage of replacing discrete proto-

cols with continuous protocols is that the latter give betterresolution of parameter setting responses, however, it is notclear if this outweighs the accuracy associated with collect-ing minutes of data at a single parameter setting. Furtherstudies are required to fully understand this tradeoff. Evalu-ating the effects of parameter settings between the actualsteps of a discrete protocol requires interpolation, whereaswith a continuous sweep a higher resolution of parametersettings within the sweep range are actually tested.When we look at the absolute reductions in metabolic

rate, as expected, reduction in metabolic rate wassmaller in Continuous-up than in Continuous-down atmost peak moment levels (Fig. 4). This finding isconsistent with the knowledge that metabolic rate fol-lows an exponential change over time, characterized inearlier studies comparing continuous and discrete stepfrequency changes [15, 22], along with exercise physiologystudies [23, 34].When we evaluated the average slope of metabolic

reduction over the range in peak moment, we found thatthe slope in Continuous-up was significantly smallerthan in Discrete but there was no significant differencebetween Continuous-down and Discrete. A possibleexplanation for the closer match between Continuous-down and Discrete can be found in the simulationprediction for the effect of physiological dynamics onContinuous-up and Continuous-down. In Discrete the

Malcolm et al. Journal of NeuroEngineering and Rehabilitation (2017) 14:72 Page 9 of 13

slope of metabolic rate versus change in peak momentwas steeper around the highest peak moment levels. Asa result of this, the effect of the physiological dynamicscan be expected to be larger at higher peak moments(Additional file 2). Since metabolic rate is at steady-stateat the very start of Continuous-down this is expected tofavorably minimize the error due to the physiologicaldynamics when the peak moment is still close to Max.(but shortly after that the error due to physiologicaldynamics will re-appear). In our study, we found thatContinuous-down by itself could potentially replaceDiscrete for measuring the relative changes in metabolicrate which would represent a time-saving of about 50%compared to Discrete. In protocols with other devices andevaluation parameters it can be hypothesized that a con-tinuous sweep protocol that starts from the steepest sideof the metabolic landscape may be the best uni-directionalsweep candidate for replacing a discrete protocol.To understand if we could completely correct for the

physiological dynamics, we explored two differentapproaches that have been previously described in theliterature. A first method was suggested by Felt et al.,2015 and consists in taking the average of measurementsobtained from increasing and decreasing parametersweeps. We applied this method by calculating Continu-ous-bidirectional, based on curve fitting on the com-bined data from Continuous-up and Continuous-down.We found that there were no significant differencesbetween Continuous-bidirectional and Discrete both inthe relative slope of metabolic reduction over the rangein peak moment and the absolute metabolic reduction atall force levels (Fig. 4). There were also no significantdifferences in reduction in metabolic rate at the dif-ferent force levels between Continuous-bidirectionaland Discrete (Fig. 4). In our protocol, Continuous-bi-directional required about the same time as Discrete(respectively 28 and 30 min).A second approach to account for the physiological

dynamics is the instantaneous metabolic rate estimationfrom Selinger et al. [22]. They showed that the change inmetabolic rate after a parameter change can be fittedwith a first-order differential equation. In an experimentinvolving step frequency optimization Felt et al. [15]proposed to use this method to find the best fitting poly-nomial of instantaneous metabolic rate from only onecontinuous sweep direction and showed that it canachieve similar accuracy for optimization as a discretestep protocol at only 1/6th of the time. In our data,applying the instantaneous metabolic rate estimationusing a fixed time constant of 42 s eliminated the signifi-cant differences between all conditions. When solvingfor the individual time constants that reduce theresponse delay between Continuous-up and Discrete andbetween Continuous-down and Discrete we found time-

constants that are much larger than the previouslyassumed time constant of 42 s. We also found that thebest fitting time constants that explain the differencebetween Continuous-up and Discrete are higher than thebest fitting time constants that explain the differencebetween Continuous-down and Discrete. This is differentfrom what has been found when comparing on-off tran-sitions in exercise physiology [35].These conclusions can be compared to recent relevant

results from Koller et al. [16]. They compared a bidirec-tional continuous sweep protocol and a discrete stepprotocol for optimizing actuation timing of an ankleexoskeleton. In their study, they evaluated the standarddeviation of the probability distribution of the estimatedoptimum actuation timing from both conditions. Theyfound a marginally higher variability in the distributionof the optimum in their continuous sweep conditionthan in their discrete step condition and attributed thisto a much shorter time that was used in their continu-ous sweep protocol. If we had found a local minimum inmetabolic rate in all conditions, we could have com-pared the validity and repeatability of both methods,similar to Koller et al. [16]. For reference the confidenceinterval analysis that was used in Koller et al. is shownin Additional file 5.The positive offset in metabolic rate as a function of

peak exosuit ankle moment in Continuous-up is anunexpected result. This results deviates from the hy-pothesis that the metabolic reduction in Discrete wouldbe midway between the metabolic reduction in Continu-ous-up and Continuous-down (Fig. 5b) [15]. There aredifferent possible explanations for the significant positiveoffset in change in metabolic reduction in Continuous-up. One possible reason could be under-fitting. Rawbreath-by-breath values from three of the seven partici-pants show a small increase in metabolic rate betweenPowered-off and Low before metabolic rate descendstoward Max. This increase in metabolic rate betweenPowered-off and Low could be noise or it could indicatethat the 4-min walking period before the continuous in-crease in peak exosuit moment started was insufficientto reach metabolic steady-state. In Discrete, Powered-offis one of five fitting points (the other four being Low toMax.) however, Powered-off is not part of the curvefitted data in Continuous-up, Continuous-down orContinuous-bidirectional. As such, the curve fits in thecontinuous conditions will be less likely to go throughzero metabolic rate at Powered-off.A limitation of all comparisons of Continuous-up,

Continuous-down and Continuous-bidirectional versusDiscrete is that Discrete is also affected by normal noiseand variability in metabolic rate and wearer adaptation.It should also be noted that our findings from compar-ing different protocols cannot be taken as general

Malcolm et al. Journal of NeuroEngineering and Rehabilitation (2017) 14:72 Page 10 of 13

guidelines, since the results dependent on the protocolmodalities [23, 25]. Longer continuous sweep conditionscould give better data resolution, but would increase therisk of slow metabolic drift [21] confounding the trend.Shorter continuous sweep conditions would reduce therisk for metabolic drift but would reduce the data reso-lution and might be more challenging for participants toadapt to and this could possibly introduce biomechanicalresponse delay. Results could have been affected by thetime to reach steady state in Discrete as well as in thesteady-state periods before Continuous-up and Continu-ous-down. While the time that we allowed should havebeen sufficient to approach the steady-state value assum-ing a time constant of 42 s the fact that we found highertime values when analyzing the individual time constantsindicates that this could have affected our results. Re-sults could also have been different if participants weregiven a shorter or longer training and warm-up time.With shorter training the relative importance of smalltime savings would be larger but it could also be possiblethat participants would be less able to quickly adapt tothe continuous conditions given less training before-hand. In future studies with discrete and continuoussweeps it could be interesting collect data from thetraining day to investigate if there is more biomechanicalresponse delay when participants first experiencechanges in exosuit assistance. While we did not find evi-dence of biomechanical response delay in kinematicsand total body kinetics it could be possible that therewould changes in muscle activation (e.g. co-contraction)that could cause and explain changes in metabolic cost.However, in this study we did not measure EMG dataand so future studies would be required to investigate ifthere was a link between changes in muscle activationand metabolic rate. For the Adjusted Continuous-up andAdjusted Continuous-down calculations it could also beinteresting to rigorously calculate and use the individualtime constant based on extra tests with a large numberof small discrete parameter changes. Other limitationsare the number of participants that were tested and thefact that we did not use correction for multiple testing.Because of all these limitations the results of this studyshould be rather interpreted as an explorative analysis ofthe possible advantages and disadvantages of discreteand continuous protocols rather than a recommendationfor either of them over the other. Finally, it should alsobe noted that aside from continuous protocols otherprotocols could provide improvements and are beingstudied such as gradient based methods [14, 15].

ConclusionDifferent groups successfully developed wearable robotsfor walking assistance, but there is still a need formethods to quickly tune actuation parameters for each

robot and population. Discrete step protocols have trad-itionally been used for evaluating responses to parameterchanges, but these are time-consuming. Recently, proto-cols have been proposed where a device parameter ischanged in a continuous way. In the present study, weshow that by continuously varying assistance magnitude(either increasing, Continuous-up; or decreasing, Con-tinuous-down) with a soft exosuit, we can achieve ashorter time and higher resolution data collection forspecific applications compared to a protocol withdiscrete step conditions (Discrete). As expected, reduc-tion in metabolic rate was smaller in Continuous-upthan in Continuous-down at most peak moment levelsdue to the physiological dynamics associated with meta-bolic measurements. When evaluating the average slopeof metabolic reduction over the entire peak momentrange there was no significant difference betweenContinuous-down and Discrete. Thus, a unidirectionalcontinuous sweep, starting from the steepest side of thelandscape of metabolic rate versus parameter change,may replace a discrete step protocol for evaluating theaverage slope of metabolic rate over peak moment.Attempting to correct for the physiological dynamics bytaking the average of Continuous-up and Continuous-down successfully removed all significant differencesversus Discrete. However, this result should not be inter-preted as an argument against the use of the adjustmentmethods from Felt et al. [15] and Selinger et al. [22]., Onthe contrary, our results also showed that the AdjustedContinuous-up and Adjusted Continuous-down correctlyestimated Discrete.For all biomechanical parameters, we found that Con-

tinuous-up and Continuous-down both closely matchedwith Discrete and therefore could replace Discrete. Sucha unidirectional sweep would save approximately halfthe time of a discrete step protocol for the methods usedin this study. Another potential advantage of replacingdiscrete step protocols with continuous sweep protocolsis that the latter can give a better resolution for param-eter setting responses in less time, however, it is not sureif or when this advantage outweighs the possible loss inaccuracy and the results will depend on the sweepduration. Evaluating the effects of parameter settingsbetween the actual steps of a discrete step protocolrequires interpolation, whereas with a continuous sweepprotocol all parameter settings within the sweep rangecan be actually tested.While this study demonstrates the promising potential

for continuous sweep protocols in studying human-robot interaction, we acknowledge that there remainssignificant work to understand how different protocolconditions (e.g. rate of change of robot parameters,study duration, uni- versus bidirectional sweeping, etc.)impact the wearer response (e.g. physiological dynamics

Malcolm et al. Journal of NeuroEngineering and Rehabilitation (2017) 14:72 Page 11 of 13

and drift and adaptation effects). However, in this study,we have shown that using continuous sweep protocolsfor examining changes in kinematic or kinetic objectiveshas potential for improving over discrete step protocols.This could be useful for certain applications wheremetabolic rate is not the prime objective or participantshave limited walking ability.

Additional files

Additional file 1: Choice of curve fitting order for fitting metabolic rateversus peak exosuit ankle moment. (PDF 97 kb)

Additional file 2: Simulation of physiological dynamics. (PDF 71 kb)

Additional file 3: Inter-stride variability [36]. (PDF 734 kb)

Additional file 4: Biomechanical delay analysis in Discrete conditiondata. (PDF 49 kb)

Additional file 5: Change in metabolic rate plotted against exosuitankle peak moment plus minus 95% confidence interval. (PDF 59 kb)

AcknowledgementsNot applicable.

FundingThis material is based upon the work supported by the Defense AdvancedResearch Projects Agency (DARPA), Warrior Web Program (Contract No.W911NF-14-C-0051), the National Science Foundation (Grant No. CNS-1446464), the National Science Foundation Graduate Research FellowshipProgram (Grant No. DGE1144152), the Samsung Scholarship, the São PauloResearch Foundation (FAPESP; Grant No. 2015/02116–1), the Robert BoschStiftung (Grant No. 32.5.G412.0003.0) and the Center for Research in HumanMovement Variability of the University of Nebraska Omaha and the NIH(P20GM109090). This work was also partially funded by the Wyss Instituteand the John A. Paulson School of Engineering and Applied Sciences atHarvard University.

Availability of data and materialsAll the data relevant to this study are presented within the manuscript.

Author’s contributionsBTQ, CJS, DMR, MG, PM, SL and CJW designed the experiment. BTQ, MGand CJW designed the high-level controller and SL developed the low-level controller. BTQ, CJS, DMR, MG, PM, SL performed the biomechanicsexperiments. BTQ, CJS, DMR, MG, PM, SL, CJW analyzed and interpretedthe data. CJS, DMR, PM, CJW prepared the manuscript. All authorsprovided critical feedback on the manuscript. All authors read andapproved the final manuscript.

Ethics approval and consent to participateAll participants provided written informed consent prior to participating inthe study. The study was approved by the Harvard Longwood Medical AreaInstitutional Review Board.

Consent for publicationNot applicable.

Competing interestsPatents have been filed with the U.S. Patent Office describing the exosuitcomponents documented in this manuscript. B.T.Q., S.L., and C.J.W. wereauthors of those patents. Harvard University has entered into a license andcollaboration agreement with ReWalk Robotics. C.J.W. is a paid consultant forReWalk Robotics.

Publisher’s NoteSpringer Nature remains neutral with regard to jurisdictional claims inpublished maps and institutional affiliations.

Author details1John A. Paulson School of Engineering and Applied Sciences, HarvardUniversity, Cambridge, MA 02138, USA. 2Wyss Institute for BiologicallyInspired Engineering, Harvard University, Cambridge, MA 02138, USA.3Department of Biomechanics and Center for Research in Human MovementVariability, University of Nebraska Omaha, Omaha, NE 68182, USA. 4Universityof São Paulo, Ribeirão Preto Medical School, Ribeirão Preto, SP, Brazil.5Department of Health Sciences and Technology, ETH Zurich, Zurich 8092,Switzerland.

Received: 21 September 2016 Accepted: 20 June 2017

References1. Malcolm P, Derave W, Galle S, De Clercq D. A simple exoskeleton that

assists plantarflexion can reduce the metabolic cost of human walking.PLoS One. 2013;8

2. Wehner M, Quinlivan B, Aubin PM, Martinez-Villapando E, Baumann M,Stirling L, et al. A lightweight soft exosuit for gait assistance. IEEE Int. Conf.Robot. Autom. Karlsruhe: IEEE; 2013. p. 3347–54.

3. Asbeck AT, De Rossi SMM, Galiana I, Ding Y, Walsh CJ. Stronger,smarter, softer: next-generation wearable robots. IEEE Robot AutomMag. 2014;21:22–33.

4. Malcolm P, Quesada RE, Caputo JM, Collins SH. The influence of push-offtiming in a robotic ankle-foot prosthesis on the energetics and mechanicsof walking. J. Neuroeng. Rehabil. 2015;12:14.

5. Jackson RW, Collins SH. An experimental comparison of the relative benefits ofwork and torque assistance in ankle exoskeletons. J Appl Physiol. 2015;119:541–57.

6. Poggensee KL, Sreenath K, Collins SH. Identifying the dynamics of a human-exoskeleton system. Columbus: Dynamic Walking; 2015. p. 2. http://www.biomechatronics.cit.cmu.edu/publications/Poggensee_2015_DW.pdf.

7. Lee S, Crea S, Malcolm P, Galiana I, Asbeck AT, Walsh CJ. ControllingNegative and Positive Power at the Ankle with a Soft Exosuit. Stockholm:IEEE; 2016. http://ieeexplore.ieee.org/abstract/document/7487531/.

8. Koller JR, Gates DH, Ferris DP, Remy CD. Confidence in the curve:establishing instantaneous cost mapping techniques using bilateral ankleexoskeletons. J Appl Physiol; 2016;2125:jap.00710.2016.

9. Awad LN, Bae J, O’Donnell K, De Rossi SMM, Hendron KL, Allen S, et al. Softrobots facilitate the restoration of gait function and efficiency after stroke.Sci Transl Med.

10. Ding Y, Panizzolo F, Siviy CJ, Malcolm P, Galiana I, Holt KG, et al. Effect oftiming of hip extension assistance during loaded walking with a softexosuit. J Neuroeng Rehabil.

11. Quinlivan BT, Sangjun L, Malcolm P, Rossi DM, Grimmer M, Siviy CJ, et al.Assistance magnitude vs. metabolic cost reductions for a tetheredmultiarticular soft exosuit. Sci Robot. 2017:2, eaah4416.

12. Caputo JM. Informing ankle-foot prosthesis design and prescription throughsystematic experimentation with a tethered robotic prosthesis. Chapter 5:User-optimal prosthesis design. 2015. http://biomechatronics.cit.cmu.edu/publications/Caputo_2015_Dissertation.pdf#page=120

13. Norris JA, Granata KP, Mitros MR, Byrne EM, Marsh AP. Effect of augmentedplantarflexion power on preferred walking speed and economy in youngand older adults. Gait Posture. 2007;25:620–7.

14. Koller JR, Gates DH, Ferris DP, Remy CD. 'Body-in-the-Loop' Optimization ofAsssistive Robotic Devices: A Validation Study. In: Hsu D, Amato N, Spring B,Jacobs S. Ann Arbor: Robotics Science and Systems (RSS); 2016. ISBN: 978-0-9923747-2-3. http://www.roboticsproceedings.org/rss12/p07.pdf.

15. Felt W, Selinger JC, Donelan JM, Remy CD. “Body-in-the-loop”: Optimizingdevice parameters using measures of instantaneous energetic cost. PLoSOne. 2015;10:e0135342.

16. Koller JR, Gates DH, Ferris DP, Remy CD. Confidence in the curve:establishing instantaneous cost mapping techniques using bilateral ankleexoskeletons. J Appl Physiol. 2016;2125:jap.00710.2016.

17. Panizzolo F, Galiana I, Asbeck AT, Siviy C, Schmidt K, Holt KG, et al. Abiologically-inspired multi-joint soft exosuit that can reduce the energy costof loaded walking. J Neuroeng Rehabil Journal of NeuroEngineering andRehabilitation. 2016;13:1–13.

18. Lamarra N, Whipp BJ, Ward SA, Wasserman K. Effect of interbreathfluctuations on characterizing exercise gas exchange kinetics. J Appl Physiol.1987;62:2003–12.

Malcolm et al. Journal of NeuroEngineering and Rehabilitation (2017) 14:72 Page 12 of 13

19. Myers J, Walsh D, Sullivan M, Froelicher V. Effect of sampling on variabilityand plateau in oxygen uptake. J Appl Physiol. 1990;68:404–10.

20. Robergs RA, Dwyer D, Astorino T. Recommendations for improved dataprocessing from expired gas analysis indirect calorimetry. Sport Med.2010;40:95–111.

21. Markovitz GH, Sayre JW, Storer TW, Cooper CB. On issues of confidence indetermining the time constant for oxygen uptake kinetics. Br J Sports Med.2004;38:553–60.

22. Selinger JC, Donelan JM. Estimating instantaneous energetic cost duringnon-steady state gait. J Appl Physiol. 2014;117:1406–15.

23. Boone J, Bourgois J. The oxygen uptake response to incremental rampexercise: methodogical and physiological issues. Sport Med. 2012;42:511–26.

24. Yandell MB, Quinlivan BT, Popov D, Walsh C, Zelik KE. Physical interfacedynamics alter how robotic exosuits augment human movement:implications for optimizing wearable assistive devices. J Neuroeng Rehabil.2017;14:40. http://jneuroengrehab.biomedcentral.com/articles/10.1186/s12984-017-0247-9.

25. Davis JA, Whipp BJ, Lamarra N, Huntsman DJ, Frank MH, Wasserman K.Effect of ramp slope on determination of aerobic parameters from the rampexercise test. Med Sci Sports Exerc. 1982;14:339–43.

26. Swanson GD, Hughson RL. On the modeling and interpretation of oxygenuptake kinetics from ramp work rate tests. J Appl Physiol. 1988;65:2453–8.

27. Gordon KE, Sawicki GS, Ferris DP. Mechanical performance of artificial pneumaticmuscles to power an ankle-foot orthosis. J Biomech. 2006;39:1832–41.

28. Caputo JM, Collins SH. Prosthetic ankle push-off work reduces metabolicrate but not collision work in non-amputee walking. Sci Rep. 2014;4:7213.

29. Brockway JM. Derivation of formulae used to calculate energy expenditurein man. Hum Nutr Clin Nutr. 1987;41:463–71.

30. Quinlivan BT, Sangjun L, Malcolm P, Rossi DM, Grimmer M. Siviy C, et al.Assistance magnitude vs metabolic cost reductions for a tetheredmultiarticular soft exosuit. 2016;4416:1–17.

31. Caputo JM, Collins SH. A Universal Ankle–Foot Prosthesis Emulator forHuman Locomotion Experiments. J Biomech Eng. 2014;136:35002.

32. Margaria R. Positive and negative work performances and their efficienciesin human locomotion. Int Z Angew Physiol. 1968;25:339–51.

33. Long AW, Roemmich RT, Bastian AJ. Blocking trial-by-trial error correctiondoes not interfere with motor learning in human walking. J Neurophysiol.2016;115:2341–8.

34. Buchfuhrer MJ, Hansen JE, Robinson TE, Sue DY, Wasserman K, Whipp BJ.Optimizing the exercise protocol for cardiopulmonary assessment. J ApplPhysiol Respir Environ Exerc Physiol. 1983;55:1558–64.

35. Paterson DH, Whipp BJ. Asymmetries of oxygen uptake transients at the on-and offset of heavy exercise in humans. J Physiol. 1991;443:575–86.

36. Hausdorff JM, Zemany L, Peng C, Goldberger AL. Maturation of gaitdynamics: stride-to-stride variability and its temporal organization inchildren. J Appl Physiol. 1999;86:1040–7.

• We accept pre-submission inquiries

• Our selector tool helps you to find the most relevant journal

• We provide round the clock customer support

• Convenient online submission

• Thorough peer review

• Inclusion in PubMed and all major indexing services

• Maximum visibility for your research

Submit your manuscript atwww.biomedcentral.com/submit

Submit your next manuscript to BioMed Central and we will help you at every step:

Malcolm et al. Journal of NeuroEngineering and Rehabilitation (2017) 14:72 Page 13 of 13